We introduce a new scaling relation between earthquake magnitude (M) and a characteristic of initial P wave displacement. By examining Japanese K‐NET data averaged in bins partitioned by Mw and hypocentral distance, we demonstrate that the P wave displacement briefly displays similar growth at the onset of rupture and that the departure time (Tdp), which is defined as the time of departure from similarity of the absolute displacement after applying a band‐pass filter, correlates with the final M in a range of 4.5 ≤ Mw ≤ 7. The scaling relation between Mw and Tdp implies that useful information on the final M can be derived while the event is still in progress because Tdp occurs before the completion of rupture. We conclude that the scaling relation is important not only for earthquake early warning but also for the source physics of earthquakes.